Key facts
- Google has released Gemini 3.8 Flash, its third "Flash" model in six weeks.
- Gemini 3.8 Flash ranks first on the DeepSWE leaderboard for solving complex software engineering problems.
- The new model demonstrates improved performance in agentic computer use tests compared to its predecessor.
- Gemini 3.8 Flash Cyber shows substantial improvements in identifying vulnerabilities and issuing patches.
- The Chrome security team reported a 2.6x increase in patch accuracy using Gemini 3.8 Flash Cyber.
- Gemini 3.8 Flash is available within the Google ecosystem for users with Pro or Ultra subscriptions.
Google has launched Gemini 3.8 Flash, its third "Flash" model in six weeks, positioning it at the top of the DeepSWE leaderboard for complex software engineering problems. This new model reportedly achieves this performance at a lower cost.
While Gemini 3.8 Flash shows an improvement over its predecessor, 3.7 Flash, in agentic computer use tests (OSWorld-2.0), it still trails behind market leader Claude Opus in this specific benchmark. GPT also struggles in this particular test.
The specialized Gemini 3.8 Flash Cyber model, replacing the 3.5 version, has demonstrated significant advancements in internal testing for cybersecurity tasks. Google claims it identifies more vulnerabilities and issues working patches more frequently. The Chrome security team noted a 2.6x increase in patch accuracy, and the Cloud team reported the model found a critical vulnerability within two hours. Partners like Wiz and Palo Alto Networks have also provided statements endorsing Gemini 3.8 Flash Cyber.
Gemini 3.8 Flash will be integrated across the Google ecosystem for users with Pro or Ultra subscriptions. However, Gemini 3.8 Flash Cyber is currently restricted to trusted testers and governments. Users can experiment with the models for free via AI Studio.